How Organizations Use AI: Evidence from ChatGPT [pdf]
OpenAI’s new paper on how organizations use ChatGPT is prompting mixed reactions, with many criticizing its structure, placement of figures, and light analytical depth despite interesting usage data. Commenters question whether it functions more as marketing than research, point to the lack of clear, measurable ROI for enterprises, and highlight the gap between top‑down corporate adoption and grassroots use by individual professionals such as teachers. Others note methodological concerns and warn that measuring message volume or “intensity of use” may be a poor proxy for real economic impact.
Paper length, format, and genre
- PDF is 69 pages, but core text ~23 double-spaced pages; rest is figures, tables, references.
- Some say it’s “not that long,” others feel the supporting material adds little.
- Debate over formatting: single-column, large line spacing, figures collected at the end.
- Several posters note this is standard for econ / social science working papers, not a polished “white paper.”
Perceived quality and rigor of the study
- Critiques that it feels like a “data dump” with limited meaningful analysis.
- Complaints that key figures are far from where they’re referenced, making reading and cross-checking claims harder.
- Some describe it as typical preprint quality; others say it “sucks,” especially the figures.
Marketing, hype, and motives
- Suspicion that frontier AI labs use such papers as stakeholder marketing rather than pure research.
- Comparisons to crypto/NFT-style whitepapers: reads like an ad encouraging adoption to “figure out what it’s good for.”
- Pushback that some terms (e.g., “general purpose technology”) are standard economics jargon, not pure hype.
Real-world AI usage in organizations and schools
- Observations that large firms may have more formal adoption due to compliance controls.
- Multiple anecdotes of teachers using ChatGPT personally to create lesson plans, slides, games, and analyze test data, with positive classroom impact.
- Others report teachers who see classroom AI as dishonest or akin to plagiarism and refuse to use it.
Education, inequality, and technology
- Long subthread on how tech (including AI) in public schools may erode analytical skills, contrasted with elite private schools emphasizing traditional, low-tech, rigorous education.
- Counterexamples where public immersion programs outperform tech-heavy private programs.
- Debate over whether differences come more from funding, parental background, peers, or administration quality.
Measuring ROI and impact
- Some interpret the paper’s implicit message as: enterprises lack clear, measurable ROI; adoption is early and metrics immature.
- Questions raised about why vendors don’t better articulate ROI if they sell “intelligence,” versus view that measurement is the customer’s job.
- One commenter warns that usage intensity is a weak proxy for economic impact, especially when juniors use AI more than seniors.
Methodological concerns
- Critique that adopters are sampled but non-adopters use full Compustat population, with no disclosed sampling rate.
- Concern this is statistically risky and may not account for small pilot deployments.